FuzzyGSSL: An Improved Graph-based Semi-Supervised Learning Through Fuzzy Label Propagation

The field of machine learning has witnessed significant advancements in semi-supervised learning (SSL).The scarcity and expense of labeled data pose a major challenge in real-world machinelearning applications.SSL offers a solution by integrating both labeled and unlabeled data for learning, bridging the gap between limited labeled data and abundant, inexpensive unlabeled data.Within this field, graph-based semi-supervised learning (GBSSL) is a prominent technique.However, GBSSL suffers from a limitation: label inference, the process of assigning pseudo-labels to unlabeled data, is fixed during graph generation.This rigidity makes it difficult to account for uncertainty in pseudo-label prediction.This paper addresses this limitation by proposing a novel approach called Fuzzy Label Propagation (FuzzyGSSL).FuzzyGSSL integrates GB-SSL with a Fuzzy Divide and Conquer Strategy.By incorporating fuzzy logic, FuzzyGSSL can account for the inherent uncertainty in pseudo-label prediction, leading to more robust and accurate results.To evaluate the effectiveness of FuzzyGSSL, the datasets were systematically divided into three ratios for further analysis.The results demonstrate that the proposed FuzzyGSSL methodology significantly outperforms traditional GBSSL in predicting pseudo labels for unlabeled data.Additionally, a detailed comparison of the outcomes between this research and the traditional approach is presented.